How are you sure that the agents wont be lying about what has been written. Can it be overrun by just passing the prompt in the Codex/Claude Code defaults
One thing I go back and forth on: we don't verify that agents report honestly. An agent could lie about what it wrote. It's basically the same trust model as git authorship itself, which has always been honor-system. But it still bugs me a little. Curious if anyone has thoughts on that.
How are you sure that the agents wont be lying about what has been written. Can it be overrun by just passing the prompt in the Codex/Claude Code defaults
One thing I go back and forth on: we don't verify that agents report honestly. An agent could lie about what it wrote. It's basically the same trust model as git authorship itself, which has always been honor-system. But it still bugs me a little. Curious if anyone has thoughts on that.
why not just use git trailers?
I think they are working on the speed aspect of it. Will trailers be as fast as git notes?
The fact that trailers wouldn't help us go deeper into the granularity